4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
8 #include <linux/config.h>
10 #include <linux/hugetlb.h>
11 #include <linux/mman.h>
12 #include <linux/slab.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/swap.h>
15 #include <linux/vmalloc.h>
16 #include <linux/pagemap.h>
17 #include <linux/namei.h>
18 #include <linux/shm.h>
19 #include <linux/blkdev.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/module.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/syscalls.h>
30 #include <asm/pgtable.h>
31 #include <asm/tlbflush.h>
32 #include <linux/swapops.h>
33 #include <linux/vs_memory.h>
35 DEFINE_SPINLOCK(swaplock);
36 unsigned int nr_swapfiles;
37 long total_swap_pages;
38 static int swap_overflow;
40 EXPORT_SYMBOL(total_swap_pages);
42 static const char Bad_file[] = "Bad swap file entry ";
43 static const char Unused_file[] = "Unused swap file entry ";
44 static const char Bad_offset[] = "Bad swap offset entry ";
45 static const char Unused_offset[] = "Unused swap offset entry ";
47 struct swap_list_t swap_list = {-1, -1};
49 struct swap_info_struct swap_info[MAX_SWAPFILES];
51 static DECLARE_MUTEX(swapon_sem);
54 * We need this because the bdev->unplug_fn can sleep and we cannot
55 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock
56 * cannot be turned into a semaphore.
58 static DECLARE_RWSEM(swap_unplug_sem);
60 #define SWAPFILE_CLUSTER 256
62 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
66 down_read(&swap_unplug_sem);
67 entry.val = page->private;
68 if (PageSwapCache(page)) {
69 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
70 struct backing_dev_info *bdi;
73 * If the page is removed from swapcache from under us (with a
74 * racy try_to_unuse/swapoff) we need an additional reference
75 * count to avoid reading garbage from page->private above. If
76 * the WARN_ON triggers during a swapoff it maybe the race
77 * condition and it's harmless. However if it triggers without
78 * swapoff it signals a problem.
80 WARN_ON(page_count(page) <= 1);
82 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
83 blk_run_backing_dev(bdi, page);
85 up_read(&swap_unplug_sem);
88 static inline int scan_swap_map(struct swap_info_struct *si)
92 * We try to cluster swap pages by allocating them
93 * sequentially in swap. Once we've allocated
94 * SWAPFILE_CLUSTER pages this way, however, we resort to
95 * first-free allocation, starting a new cluster. This
96 * prevents us from scattering swap pages all over the entire
97 * swap partition, so that we reduce overall disk seek times
98 * between swap pages. -- sct */
100 while (si->cluster_next <= si->highest_bit) {
101 offset = si->cluster_next++;
102 if (si->swap_map[offset])
108 si->cluster_nr = SWAPFILE_CLUSTER;
110 /* try to find an empty (even not aligned) cluster. */
111 offset = si->lowest_bit;
113 if (offset+SWAPFILE_CLUSTER-1 <= si->highest_bit)
116 for (nr = offset; nr < offset+SWAPFILE_CLUSTER; nr++)
117 if (si->swap_map[nr])
120 goto check_next_cluster;
122 /* We found a completly empty cluster, so start
127 /* No luck, so now go finegrined as usual. -Andrea */
128 for (offset = si->lowest_bit; offset <= si->highest_bit ; offset++) {
129 if (si->swap_map[offset])
131 si->lowest_bit = offset+1;
133 if (offset == si->lowest_bit)
135 if (offset == si->highest_bit)
137 if (si->lowest_bit > si->highest_bit) {
138 si->lowest_bit = si->max;
141 si->swap_map[offset] = 1;
144 si->cluster_next = offset+1;
147 si->lowest_bit = si->max;
152 swp_entry_t get_swap_page(void)
154 struct swap_info_struct * p;
155 unsigned long offset;
157 int type, wrapped = 0;
159 entry.val = 0; /* Out of memory */
161 type = swap_list.next;
164 if (nr_swap_pages <= 0)
168 p = &swap_info[type];
169 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
171 offset = scan_swap_map(p);
172 swap_device_unlock(p);
174 entry = swp_entry(type,offset);
175 type = swap_info[type].next;
177 p->prio != swap_info[type].prio) {
178 swap_list.next = swap_list.head;
180 swap_list.next = type;
187 if (type < 0 || p->prio != swap_info[type].prio) {
188 type = swap_list.head;
193 goto out; /* out of swap space */
200 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
202 struct swap_info_struct * p;
203 unsigned long offset, type;
207 type = swp_type(entry);
208 if (type >= nr_swapfiles)
210 p = & swap_info[type];
211 if (!(p->flags & SWP_USED))
213 offset = swp_offset(entry);
214 if (offset >= p->max)
216 if (!p->swap_map[offset])
219 if (p->prio > swap_info[swap_list.next].prio)
220 swap_list.next = type;
225 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
228 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
231 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
234 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
239 static void swap_info_put(struct swap_info_struct * p)
241 swap_device_unlock(p);
245 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
247 int count = p->swap_map[offset];
249 if (count < SWAP_MAP_MAX) {
251 p->swap_map[offset] = count;
253 if (offset < p->lowest_bit)
254 p->lowest_bit = offset;
255 if (offset > p->highest_bit)
256 p->highest_bit = offset;
265 * Caller has made sure that the swapdevice corresponding to entry
266 * is still around or has not been recycled.
268 void swap_free(swp_entry_t entry)
270 struct swap_info_struct * p;
272 p = swap_info_get(entry);
274 swap_entry_free(p, swp_offset(entry));
280 * Check if we're the only user of a swap page,
281 * when the page is locked.
283 static int exclusive_swap_page(struct page *page)
286 struct swap_info_struct * p;
289 entry.val = page->private;
290 p = swap_info_get(entry);
292 /* Is the only swap cache user the cache itself? */
293 if (p->swap_map[swp_offset(entry)] == 1) {
294 /* Recheck the page count with the swapcache lock held.. */
295 write_lock_irq(&swapper_space.tree_lock);
296 if (page_count(page) == 2)
298 write_unlock_irq(&swapper_space.tree_lock);
306 * We can use this swap cache entry directly
307 * if there are no other references to it.
309 * Here "exclusive_swap_page()" does the real
310 * work, but we opportunistically check whether
311 * we need to get all the locks first..
313 int can_share_swap_page(struct page *page)
317 if (!PageLocked(page))
319 switch (page_count(page)) {
321 if (!PagePrivate(page))
325 if (!PageSwapCache(page))
327 retval = exclusive_swap_page(page);
330 if (PageReserved(page))
338 * Work out if there are any other processes sharing this
339 * swap cache page. Free it if you can. Return success.
341 int remove_exclusive_swap_page(struct page *page)
344 struct swap_info_struct * p;
347 BUG_ON(PagePrivate(page));
348 BUG_ON(!PageLocked(page));
350 if (!PageSwapCache(page))
352 if (PageWriteback(page))
354 if (page_count(page) != 2) /* 2: us + cache */
357 entry.val = page->private;
358 p = swap_info_get(entry);
362 /* Is the only swap cache user the cache itself? */
364 if (p->swap_map[swp_offset(entry)] == 1) {
365 /* Recheck the page count with the swapcache lock held.. */
366 write_lock_irq(&swapper_space.tree_lock);
367 if ((page_count(page) == 2) && !PageWriteback(page)) {
368 __delete_from_swap_cache(page);
372 write_unlock_irq(&swapper_space.tree_lock);
378 page_cache_release(page);
385 * Free the swap entry like above, but also try to
386 * free the page cache entry if it is the last user.
388 void free_swap_and_cache(swp_entry_t entry)
390 struct swap_info_struct * p;
391 struct page *page = NULL;
393 p = swap_info_get(entry);
395 if (swap_entry_free(p, swp_offset(entry)) == 1)
396 page = find_trylock_page(&swapper_space, entry.val);
402 BUG_ON(PagePrivate(page));
403 page_cache_get(page);
404 one_user = (page_count(page) == 2);
405 /* Only cache user (+us), or swap space full? Free it! */
406 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
407 delete_from_swap_cache(page);
411 page_cache_release(page);
416 * Always set the resulting pte to be nowrite (the same as COW pages
417 * after one process has exited). We don't know just how many PTEs will
418 * share this swap entry, so be cautious and let do_wp_page work out
419 * what to do if a write is requested later.
421 * vma->vm_mm->page_table_lock is held.
423 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
424 unsigned long addr, swp_entry_t entry, struct page *page)
426 inc_mm_counter(vma->vm_mm, rss);
428 set_pte_at(vma->vm_mm, addr, pte,
429 pte_mkold(mk_pte(page, vma->vm_page_prot)));
430 page_add_anon_rmap(page, vma, addr);
433 * Move the page to the active list so it is not
434 * immediately swapped out again after swapon.
439 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
440 unsigned long addr, unsigned long end,
441 swp_entry_t entry, struct page *page)
444 pte_t swp_pte = swp_entry_to_pte(entry);
446 pte = pte_offset_map(pmd, addr);
449 * swapoff spends a _lot_ of time in this loop!
450 * Test inline before going to call unuse_pte.
452 if (unlikely(pte_same(*pte, swp_pte))) {
453 unuse_pte(vma, pte, addr, entry, page);
457 } while (pte++, addr += PAGE_SIZE, addr != end);
462 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
463 unsigned long addr, unsigned long end,
464 swp_entry_t entry, struct page *page)
469 pmd = pmd_offset(pud, addr);
471 next = pmd_addr_end(addr, end);
472 if (pmd_none_or_clear_bad(pmd))
474 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
476 } while (pmd++, addr = next, addr != end);
480 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
481 unsigned long addr, unsigned long end,
482 swp_entry_t entry, struct page *page)
487 pud = pud_offset(pgd, addr);
489 next = pud_addr_end(addr, end);
490 if (pud_none_or_clear_bad(pud))
492 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
494 } while (pud++, addr = next, addr != end);
498 static int unuse_vma(struct vm_area_struct *vma,
499 swp_entry_t entry, struct page *page)
502 unsigned long addr, end, next;
505 addr = page_address_in_vma(page, vma);
509 end = addr + PAGE_SIZE;
511 addr = vma->vm_start;
515 pgd = pgd_offset(vma->vm_mm, addr);
517 next = pgd_addr_end(addr, end);
518 if (pgd_none_or_clear_bad(pgd))
520 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
522 } while (pgd++, addr = next, addr != end);
526 static int unuse_mm(struct mm_struct *mm,
527 swp_entry_t entry, struct page *page)
529 struct vm_area_struct *vma;
531 if (!down_read_trylock(&mm->mmap_sem)) {
533 * Our reference to the page stops try_to_unmap_one from
534 * unmapping its ptes, so swapoff can make progress.
537 down_read(&mm->mmap_sem);
540 spin_lock(&mm->page_table_lock);
541 for (vma = mm->mmap; vma; vma = vma->vm_next) {
542 if (vma->anon_vma && unuse_vma(vma, entry, page))
545 spin_unlock(&mm->page_table_lock);
546 up_read(&mm->mmap_sem);
548 * Currently unuse_mm cannot fail, but leave error handling
549 * at call sites for now, since we change it from time to time.
555 * Scan swap_map from current position to next entry still in use.
556 * Recycle to start on reaching the end, returning 0 when empty.
558 static int find_next_to_unuse(struct swap_info_struct *si, int prev)
565 * No need for swap_device_lock(si) here: we're just looking
566 * for whether an entry is in use, not modifying it; false
567 * hits are okay, and sys_swapoff() has already prevented new
568 * allocations from this area (while holding swap_list_lock()).
577 * No entries in use at top of swap_map,
578 * loop back to start and recheck there.
584 count = si->swap_map[i];
585 if (count && count != SWAP_MAP_BAD)
592 * We completely avoid races by reading each swap page in advance,
593 * and then search for the process using it. All the necessary
594 * page table adjustments can then be made atomically.
596 static int try_to_unuse(unsigned int type)
598 struct swap_info_struct * si = &swap_info[type];
599 struct mm_struct *start_mm;
600 unsigned short *swap_map;
601 unsigned short swcount;
606 int reset_overflow = 0;
610 * When searching mms for an entry, a good strategy is to
611 * start at the first mm we freed the previous entry from
612 * (though actually we don't notice whether we or coincidence
613 * freed the entry). Initialize this start_mm with a hold.
615 * A simpler strategy would be to start at the last mm we
616 * freed the previous entry from; but that would take less
617 * advantage of mmlist ordering, which clusters forked mms
618 * together, child after parent. If we race with dup_mmap(), we
619 * prefer to resolve parent before child, lest we miss entries
620 * duplicated after we scanned child: using last mm would invert
621 * that. Though it's only a serious concern when an overflowed
622 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
625 atomic_inc(&init_mm.mm_users);
628 * Keep on scanning until all entries have gone. Usually,
629 * one pass through swap_map is enough, but not necessarily:
630 * there are races when an instance of an entry might be missed.
632 while ((i = find_next_to_unuse(si, i)) != 0) {
633 if (signal_pending(current)) {
639 * Get a page for the entry, using the existing swap
640 * cache page if there is one. Otherwise, get a clean
641 * page and read the swap into it.
643 swap_map = &si->swap_map[i];
644 entry = swp_entry(type, i);
645 page = read_swap_cache_async(entry, NULL, 0);
648 * Either swap_duplicate() failed because entry
649 * has been freed independently, and will not be
650 * reused since sys_swapoff() already disabled
651 * allocation from here, or alloc_page() failed.
660 * Don't hold on to start_mm if it looks like exiting.
662 if (atomic_read(&start_mm->mm_users) == 1) {
665 atomic_inc(&init_mm.mm_users);
669 * Wait for and lock page. When do_swap_page races with
670 * try_to_unuse, do_swap_page can handle the fault much
671 * faster than try_to_unuse can locate the entry. This
672 * apparently redundant "wait_on_page_locked" lets try_to_unuse
673 * defer to do_swap_page in such a case - in some tests,
674 * do_swap_page and try_to_unuse repeatedly compete.
676 wait_on_page_locked(page);
677 wait_on_page_writeback(page);
679 wait_on_page_writeback(page);
682 * Remove all references to entry.
683 * Whenever we reach init_mm, there's no address space
684 * to search, but use it as a reminder to search shmem.
689 if (start_mm == &init_mm)
690 shmem = shmem_unuse(entry, page);
692 retval = unuse_mm(start_mm, entry, page);
695 int set_start_mm = (*swap_map >= swcount);
696 struct list_head *p = &start_mm->mmlist;
697 struct mm_struct *new_start_mm = start_mm;
698 struct mm_struct *prev_mm = start_mm;
699 struct mm_struct *mm;
701 atomic_inc(&new_start_mm->mm_users);
702 atomic_inc(&prev_mm->mm_users);
703 spin_lock(&mmlist_lock);
704 while (*swap_map > 1 && !retval &&
705 (p = p->next) != &start_mm->mmlist) {
706 mm = list_entry(p, struct mm_struct, mmlist);
707 if (atomic_inc_return(&mm->mm_users) == 1) {
708 atomic_dec(&mm->mm_users);
711 spin_unlock(&mmlist_lock);
720 else if (mm == &init_mm) {
722 shmem = shmem_unuse(entry, page);
724 retval = unuse_mm(mm, entry, page);
725 if (set_start_mm && *swap_map < swcount) {
727 atomic_inc(&mm->mm_users);
731 spin_lock(&mmlist_lock);
733 spin_unlock(&mmlist_lock);
736 start_mm = new_start_mm;
740 page_cache_release(page);
745 * How could swap count reach 0x7fff when the maximum
746 * pid is 0x7fff, and there's no way to repeat a swap
747 * page within an mm (except in shmem, where it's the
748 * shared object which takes the reference count)?
749 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
751 * If that's wrong, then we should worry more about
752 * exit_mmap() and do_munmap() cases described above:
753 * we might be resetting SWAP_MAP_MAX too early here.
754 * We know "Undead"s can happen, they're okay, so don't
755 * report them; but do report if we reset SWAP_MAP_MAX.
757 if (*swap_map == SWAP_MAP_MAX) {
758 swap_device_lock(si);
760 swap_device_unlock(si);
765 * If a reference remains (rare), we would like to leave
766 * the page in the swap cache; but try_to_unmap could
767 * then re-duplicate the entry once we drop page lock,
768 * so we might loop indefinitely; also, that page could
769 * not be swapped out to other storage meanwhile. So:
770 * delete from cache even if there's another reference,
771 * after ensuring that the data has been saved to disk -
772 * since if the reference remains (rarer), it will be
773 * read from disk into another page. Splitting into two
774 * pages would be incorrect if swap supported "shared
775 * private" pages, but they are handled by tmpfs files.
777 * Note shmem_unuse already deleted a swappage from
778 * the swap cache, unless the move to filepage failed:
779 * in which case it left swappage in cache, lowered its
780 * swap count to pass quickly through the loops above,
781 * and now we must reincrement count to try again later.
783 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
784 struct writeback_control wbc = {
785 .sync_mode = WB_SYNC_NONE,
788 swap_writepage(page, &wbc);
790 wait_on_page_writeback(page);
792 if (PageSwapCache(page)) {
794 swap_duplicate(entry);
796 delete_from_swap_cache(page);
800 * So we could skip searching mms once swap count went
801 * to 1, we did not mark any present ptes as dirty: must
802 * mark page dirty so shrink_list will preserve it.
806 page_cache_release(page);
809 * Make sure that we aren't completely killing
810 * interactive performance.
816 if (reset_overflow) {
817 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
824 * After a successful try_to_unuse, if no swap is now in use, we know we
825 * can empty the mmlist. swap_list_lock must be held on entry and exit.
826 * Note that mmlist_lock nests inside swap_list_lock, and an mm must be
827 * added to the mmlist just after page_duplicate - before would be racy.
829 static void drain_mmlist(void)
831 struct list_head *p, *next;
834 for (i = 0; i < nr_swapfiles; i++)
835 if (swap_info[i].inuse_pages)
837 spin_lock(&mmlist_lock);
838 list_for_each_safe(p, next, &init_mm.mmlist)
840 spin_unlock(&mmlist_lock);
844 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
845 * corresponds to page offset `offset'.
847 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
849 struct swap_extent *se = sis->curr_swap_extent;
850 struct swap_extent *start_se = se;
853 struct list_head *lh;
855 if (se->start_page <= offset &&
856 offset < (se->start_page + se->nr_pages)) {
857 return se->start_block + (offset - se->start_page);
860 if (lh == &sis->extent_list)
862 se = list_entry(lh, struct swap_extent, list);
863 sis->curr_swap_extent = se;
864 BUG_ON(se == start_se); /* It *must* be present */
869 * Free all of a swapdev's extent information
871 static void destroy_swap_extents(struct swap_info_struct *sis)
873 while (!list_empty(&sis->extent_list)) {
874 struct swap_extent *se;
876 se = list_entry(sis->extent_list.next,
877 struct swap_extent, list);
885 * Add a block range (and the corresponding page range) into this swapdev's
886 * extent list. The extent list is kept sorted in block order.
888 * This function rather assumes that it is called in ascending sector_t order.
889 * It doesn't look for extent coalescing opportunities.
892 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
893 unsigned long nr_pages, sector_t start_block)
895 struct swap_extent *se;
896 struct swap_extent *new_se;
897 struct list_head *lh;
899 lh = sis->extent_list.next; /* The highest-addressed block */
900 while (lh != &sis->extent_list) {
901 se = list_entry(lh, struct swap_extent, list);
902 if (se->start_block + se->nr_pages == start_block &&
903 se->start_page + se->nr_pages == start_page) {
905 se->nr_pages += nr_pages;
912 * No merge. Insert a new extent, preserving ordering.
914 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
917 new_se->start_page = start_page;
918 new_se->nr_pages = nr_pages;
919 new_se->start_block = start_block;
921 lh = sis->extent_list.prev; /* The lowest block */
922 while (lh != &sis->extent_list) {
923 se = list_entry(lh, struct swap_extent, list);
924 if (se->start_block > start_block)
928 list_add_tail(&new_se->list, lh);
934 * A `swap extent' is a simple thing which maps a contiguous range of pages
935 * onto a contiguous range of disk blocks. An ordered list of swap extents
936 * is built at swapon time and is then used at swap_writepage/swap_readpage
937 * time for locating where on disk a page belongs.
939 * If the swapfile is an S_ISBLK block device, a single extent is installed.
940 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
941 * swap files identically.
943 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
944 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
945 * swapfiles are handled *identically* after swapon time.
947 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
948 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
949 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
950 * requirements, they are simply tossed out - we will never use those blocks
953 * For S_ISREG swapfiles we hold i_sem across the life of the swapon. This
954 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
955 * which will scribble on the fs.
957 * The amount of disk space which a single swap extent represents varies.
958 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
959 * extents in the list. To avoid much list walking, we cache the previous
960 * search location in `curr_swap_extent', and start new searches from there.
961 * This is extremely effective. The average number of iterations in
962 * map_swap_page() has been measured at about 0.3 per page. - akpm.
964 static int setup_swap_extents(struct swap_info_struct *sis)
967 unsigned blocks_per_page;
968 unsigned long page_no;
970 sector_t probe_block;
974 inode = sis->swap_file->f_mapping->host;
975 if (S_ISBLK(inode->i_mode)) {
976 ret = add_swap_extent(sis, 0, sis->max, 0);
980 blkbits = inode->i_blkbits;
981 blocks_per_page = PAGE_SIZE >> blkbits;
984 * Map all the blocks into the extent list. This code doesn't try
989 last_block = i_size_read(inode) >> blkbits;
990 while ((probe_block + blocks_per_page) <= last_block &&
991 page_no < sis->max) {
992 unsigned block_in_page;
993 sector_t first_block;
995 first_block = bmap(inode, probe_block);
996 if (first_block == 0)
1000 * It must be PAGE_SIZE aligned on-disk
1002 if (first_block & (blocks_per_page - 1)) {
1007 for (block_in_page = 1; block_in_page < blocks_per_page;
1011 block = bmap(inode, probe_block + block_in_page);
1014 if (block != first_block + block_in_page) {
1022 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1024 ret = add_swap_extent(sis, page_no, 1,
1025 first_block >> (PAGE_SHIFT - blkbits));
1029 probe_block += blocks_per_page;
1037 sis->highest_bit = page_no - 1;
1039 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1040 struct swap_extent, list);
1043 printk(KERN_ERR "swapon: swapfile has holes\n");
1049 #if 0 /* We don't need this yet */
1050 #include <linux/backing-dev.h>
1051 int page_queue_congested(struct page *page)
1053 struct backing_dev_info *bdi;
1055 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1057 if (PageSwapCache(page)) {
1058 swp_entry_t entry = { .val = page->private };
1059 struct swap_info_struct *sis;
1061 sis = get_swap_info_struct(swp_type(entry));
1062 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1064 bdi = page->mapping->backing_dev_info;
1065 return bdi_write_congested(bdi);
1069 asmlinkage long sys_swapoff(const char __user * specialfile)
1071 struct swap_info_struct * p = NULL;
1072 unsigned short *swap_map;
1073 struct file *swap_file, *victim;
1074 struct address_space *mapping;
1075 struct inode *inode;
1080 if (!capable(CAP_SYS_ADMIN))
1083 pathname = getname(specialfile);
1084 err = PTR_ERR(pathname);
1085 if (IS_ERR(pathname))
1088 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1090 err = PTR_ERR(victim);
1094 mapping = victim->f_mapping;
1097 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1098 p = swap_info + type;
1099 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1100 if (p->swap_file->f_mapping == mapping)
1110 if (!security_vm_enough_memory(p->pages))
1111 vm_unacct_memory(p->pages);
1118 swap_list.head = p->next;
1120 swap_info[prev].next = p->next;
1122 if (type == swap_list.next) {
1123 /* just pick something that's safe... */
1124 swap_list.next = swap_list.head;
1126 nr_swap_pages -= p->pages;
1127 total_swap_pages -= p->pages;
1128 p->flags &= ~SWP_WRITEOK;
1130 current->flags |= PF_SWAPOFF;
1131 err = try_to_unuse(type);
1132 current->flags &= ~PF_SWAPOFF;
1134 /* wait for any unplug function to finish */
1135 down_write(&swap_unplug_sem);
1136 up_write(&swap_unplug_sem);
1139 /* re-insert swap space back into swap_list */
1141 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1142 if (p->prio >= swap_info[i].prio)
1146 swap_list.head = swap_list.next = p - swap_info;
1148 swap_info[prev].next = p - swap_info;
1149 nr_swap_pages += p->pages;
1150 total_swap_pages += p->pages;
1151 p->flags |= SWP_WRITEOK;
1158 swap_device_lock(p);
1159 swap_file = p->swap_file;
1160 p->swap_file = NULL;
1162 swap_map = p->swap_map;
1165 destroy_swap_extents(p);
1166 swap_device_unlock(p);
1170 inode = mapping->host;
1171 if (S_ISBLK(inode->i_mode)) {
1172 struct block_device *bdev = I_BDEV(inode);
1173 set_blocksize(bdev, p->old_block_size);
1176 down(&inode->i_sem);
1177 inode->i_flags &= ~S_SWAPFILE;
1180 filp_close(swap_file, NULL);
1184 filp_close(victim, NULL);
1189 #ifdef CONFIG_PROC_FS
1191 static void *swap_start(struct seq_file *swap, loff_t *pos)
1193 struct swap_info_struct *ptr = swap_info;
1199 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1200 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1209 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1211 struct swap_info_struct *ptr = v;
1212 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1214 for (++ptr; ptr < endptr; ptr++) {
1215 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1224 static void swap_stop(struct seq_file *swap, void *v)
1229 static int swap_show(struct seq_file *swap, void *v)
1231 struct swap_info_struct *ptr = v;
1236 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1238 file = ptr->swap_file;
1239 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1240 seq_printf(swap, "%*s%s\t%d\t%ld\t%d\n",
1241 len < 40 ? 40 - len : 1, " ",
1242 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1243 "partition" : "file\t",
1244 ptr->pages << (PAGE_SHIFT - 10),
1245 ptr->inuse_pages << (PAGE_SHIFT - 10),
1250 static struct seq_operations swaps_op = {
1251 .start = swap_start,
1257 static int swaps_open(struct inode *inode, struct file *file)
1259 return seq_open(file, &swaps_op);
1262 static struct file_operations proc_swaps_operations = {
1265 .llseek = seq_lseek,
1266 .release = seq_release,
1269 static int __init procswaps_init(void)
1271 struct proc_dir_entry *entry;
1273 entry = create_proc_entry("swaps", 0, NULL);
1275 entry->proc_fops = &proc_swaps_operations;
1278 __initcall(procswaps_init);
1279 #endif /* CONFIG_PROC_FS */
1282 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1284 * The swapon system call
1286 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1288 struct swap_info_struct * p;
1290 struct block_device *bdev = NULL;
1291 struct file *swap_file = NULL;
1292 struct address_space *mapping;
1296 static int least_priority;
1297 union swap_header *swap_header = NULL;
1298 int swap_header_version;
1299 int nr_good_pages = 0;
1300 unsigned long maxpages = 1;
1302 unsigned short *swap_map;
1303 struct page *page = NULL;
1304 struct inode *inode = NULL;
1307 if (!capable(CAP_SYS_ADMIN))
1311 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1312 if (!(p->flags & SWP_USED))
1316 * Test if adding another swap device is possible. There are
1317 * two limiting factors: 1) the number of bits for the swap
1318 * type swp_entry_t definition and 2) the number of bits for
1319 * the swap type in the swap ptes as defined by the different
1320 * architectures. To honor both limitations a swap entry
1321 * with swap offset 0 and swap type ~0UL is created, encoded
1322 * to a swap pte, decoded to a swp_entry_t again and finally
1323 * the swap type part is extracted. This will mask all bits
1324 * from the initial ~0UL that can't be encoded in either the
1325 * swp_entry_t or the architecture definition of a swap pte.
1327 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1331 if (type >= nr_swapfiles)
1332 nr_swapfiles = type+1;
1333 INIT_LIST_HEAD(&p->extent_list);
1334 p->flags = SWP_USED;
1336 p->swap_file = NULL;
1337 p->old_block_size = 0;
1343 spin_lock_init(&p->sdev_lock);
1345 if (swap_flags & SWAP_FLAG_PREFER) {
1347 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1349 p->prio = --least_priority;
1352 name = getname(specialfile);
1353 error = PTR_ERR(name);
1358 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1359 error = PTR_ERR(swap_file);
1360 if (IS_ERR(swap_file)) {
1365 p->swap_file = swap_file;
1366 mapping = swap_file->f_mapping;
1367 inode = mapping->host;
1370 for (i = 0; i < nr_swapfiles; i++) {
1371 struct swap_info_struct *q = &swap_info[i];
1373 if (i == type || !q->swap_file)
1375 if (mapping == q->swap_file->f_mapping)
1380 if (S_ISBLK(inode->i_mode)) {
1381 bdev = I_BDEV(inode);
1382 error = bd_claim(bdev, sys_swapon);
1387 p->old_block_size = block_size(bdev);
1388 error = set_blocksize(bdev, PAGE_SIZE);
1392 } else if (S_ISREG(inode->i_mode)) {
1393 p->bdev = inode->i_sb->s_bdev;
1394 down(&inode->i_sem);
1396 if (IS_SWAPFILE(inode)) {
1404 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1407 * Read the swap header.
1409 if (!mapping->a_ops->readpage) {
1413 page = read_cache_page(mapping, 0,
1414 (filler_t *)mapping->a_ops->readpage, swap_file);
1416 error = PTR_ERR(page);
1419 wait_on_page_locked(page);
1420 if (!PageUptodate(page))
1423 swap_header = page_address(page);
1425 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1426 swap_header_version = 1;
1427 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1428 swap_header_version = 2;
1430 printk("Unable to find swap-space signature\n");
1435 switch (swap_header_version) {
1437 printk(KERN_ERR "version 0 swap is no longer supported. "
1438 "Use mkswap -v1 %s\n", name);
1442 /* Check the swap header's sub-version and the size of
1443 the swap file and bad block lists */
1444 if (swap_header->info.version != 1) {
1446 "Unable to handle swap header version %d\n",
1447 swap_header->info.version);
1454 * Find out how many pages are allowed for a single swap
1455 * device. There are two limiting factors: 1) the number of
1456 * bits for the swap offset in the swp_entry_t type and
1457 * 2) the number of bits in the a swap pte as defined by
1458 * the different architectures. In order to find the
1459 * largest possible bit mask a swap entry with swap type 0
1460 * and swap offset ~0UL is created, encoded to a swap pte,
1461 * decoded to a swp_entry_t again and finally the swap
1462 * offset is extracted. This will mask all the bits from
1463 * the initial ~0UL mask that can't be encoded in either
1464 * the swp_entry_t or the architecture definition of a
1467 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1468 if (maxpages > swap_header->info.last_page)
1469 maxpages = swap_header->info.last_page;
1470 p->highest_bit = maxpages - 1;
1473 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1476 /* OK, set up the swap map and apply the bad block list */
1477 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1483 memset(p->swap_map, 0, maxpages * sizeof(short));
1484 for (i=0; i<swap_header->info.nr_badpages; i++) {
1485 int page = swap_header->info.badpages[i];
1486 if (page <= 0 || page >= swap_header->info.last_page)
1489 p->swap_map[page] = SWAP_MAP_BAD;
1491 nr_good_pages = swap_header->info.last_page -
1492 swap_header->info.nr_badpages -
1493 1 /* header page */;
1498 if (swapfilesize && maxpages > swapfilesize) {
1500 "Swap area shorter than signature indicates\n");
1504 if (!nr_good_pages) {
1505 printk(KERN_WARNING "Empty swap-file\n");
1509 p->swap_map[0] = SWAP_MAP_BAD;
1511 p->pages = nr_good_pages;
1513 error = setup_swap_extents(p);
1519 swap_device_lock(p);
1520 p->flags = SWP_ACTIVE;
1521 nr_swap_pages += nr_good_pages;
1522 total_swap_pages += nr_good_pages;
1523 printk(KERN_INFO "Adding %dk swap on %s. Priority:%d extents:%d\n",
1524 nr_good_pages<<(PAGE_SHIFT-10), name,
1525 p->prio, p->nr_extents);
1527 /* insert swap space into swap_list: */
1529 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1530 if (p->prio >= swap_info[i].prio) {
1537 swap_list.head = swap_list.next = p - swap_info;
1539 swap_info[prev].next = p - swap_info;
1541 swap_device_unlock(p);
1548 set_blocksize(bdev, p->old_block_size);
1553 swap_map = p->swap_map;
1554 p->swap_file = NULL;
1557 if (!(swap_flags & SWAP_FLAG_PREFER))
1560 destroy_swap_extents(p);
1563 filp_close(swap_file, NULL);
1565 if (page && !IS_ERR(page)) {
1567 page_cache_release(page);
1573 inode->i_flags |= S_SWAPFILE;
1579 void si_swapinfo(struct sysinfo *val)
1582 unsigned long nr_to_be_unused = 0;
1585 for (i = 0; i < nr_swapfiles; i++) {
1586 if (!(swap_info[i].flags & SWP_USED) ||
1587 (swap_info[i].flags & SWP_WRITEOK))
1589 nr_to_be_unused += swap_info[i].inuse_pages;
1591 val->freeswap = nr_swap_pages + nr_to_be_unused;
1592 val->totalswap = total_swap_pages + nr_to_be_unused;
1594 if (vx_flags(VXF_VIRT_MEM, 0))
1595 vx_vsi_swapinfo(val);
1599 * Verify that a swap entry is valid and increment its swap map count.
1601 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1602 * "permanent", but will be reclaimed by the next swapoff.
1604 int swap_duplicate(swp_entry_t entry)
1606 struct swap_info_struct * p;
1607 unsigned long offset, type;
1610 type = swp_type(entry);
1611 if (type >= nr_swapfiles)
1613 p = type + swap_info;
1614 offset = swp_offset(entry);
1616 swap_device_lock(p);
1617 if (offset < p->max && p->swap_map[offset]) {
1618 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1619 p->swap_map[offset]++;
1621 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1622 if (swap_overflow++ < 5)
1623 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1624 p->swap_map[offset] = SWAP_MAP_MAX;
1628 swap_device_unlock(p);
1633 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1637 struct swap_info_struct *
1638 get_swap_info_struct(unsigned type)
1640 return &swap_info[type];
1644 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1645 * reference on the swaphandle, it doesn't matter if it becomes unused.
1647 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1649 int ret = 0, i = 1 << page_cluster;
1651 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1653 if (!page_cluster) /* no readahead */
1655 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1656 if (!toff) /* first page is swap header */
1660 swap_device_lock(swapdev);
1662 /* Don't read-ahead past the end of the swap area */
1663 if (toff >= swapdev->max)
1665 /* Don't read in free or bad pages */
1666 if (!swapdev->swap_map[toff])
1668 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1673 swap_device_unlock(swapdev);